This paper presents an algorithm, DCFold, that automatically predicts the common secondary structure of a set of aligned homologous RNA sequences. It is based on the comparative approach. Helices are searched in one of the sequences, called the ‘target sequence’, and compared to the helices in the other sequences, called the ‘test sequences’. Our algorithm searches in the target sequence for palindromes that have a high probability to define helices that are conserved in the test sequences. This selection of significant palindromes is based on criteria that take into account their length and their mutation rate. A recursive search of helices, starting from these likely ones, is implemented using the ‘divide and conquer’ approach. Indeed, as pseudo-knots are not searched by DCFold, a selected palindrome (p, p′) makes possible to divide the initial sequence into two sequences, the internal one and the one resulting from the concatenation of the two external ones. New palindromes can be searched independently in these subsequences. This algorithm was run on ribosomal RNA sequences and recovered very efficiently their common secondary structures.